A fold marker interrupts a reader mid-output and hands back bookkeeping. The reader is
then holding a decision they did not have before: is what is missing the part I needed,
do I pay for a retrieval, or do I just run the command again.
Three separate reports are the same defect, and each was fixed one wording at a time:
Three instances, three point fixes, no rule. The rule the three imply:
A marker states the fact the reader is holding a question about. It never states what
the ledger did.
That is a different sentence from "make the marker shorter", which is the constraint that
has driven marker wording so far because marker length gates folding (#450 measured 0.3
points on a 22 byte trim). Both matter, and the current wordings were chosen with only the
second in hand.
Why this is not just wording
The two remaining shapes to check against the rule are the run marker (N lines already shown, N lines already shown from <command>) and the project run marker
(N lines not shown here). Both state provenance. Neither says what a reader most often
wants at that point, which is whether the elided lines differ from what they already have.
Blocked on #766, deliberately
#766 adds marker_kind to ledger_folds and the retrieve-rate query. That query orders
this work by which shapes actually cost a round trip, instead of by which example someone
remembers. Three wording changes have shipped on reasoning alone; the fourth should not.
Cost to keep in view: any wording that grows the marker reduces fold volume, since the
gate weighs the rendered string. #755 traded 15 bytes on one arm for the identity, and
that trade was never measured either.
A fold marker interrupts a reader mid-output and hands back bookkeeping. The reader is
then holding a decision they did not have before: is what is missing the part I needed,
do I pay for a retrieval, or do I just run the command again.
Three separate reports are the same defect, and each was fixed one wording at a time:
N lines already shown, which a reader took as "thecommand printed nothing". Re-running a command is how a failure gets verified, so that
misreading is the expensive one. Fixed by stating the identity instead.
from an earlier session, which reads as "you have seenthis" and means the opposite. A reader acted on it, took a help page missing its
Commands:block as complete, and concluded the CLI had no uninstall.N lines already shownwhen the answer the reader wanted wasthat the value had not changed. The surviving half of the reply was an unrelated table
that read as a complete result.
Three instances, three point fixes, no rule. The rule the three imply:
That is a different sentence from "make the marker shorter", which is the constraint that
has driven marker wording so far because marker length gates folding (#450 measured 0.3
points on a 22 byte trim). Both matter, and the current wordings were chosen with only the
second in hand.
Why this is not just wording
The two remaining shapes to check against the rule are the run marker (
N lines already shown,N lines already shown from <command>) and the project run marker(
N lines not shown here). Both state provenance. Neither says what a reader most oftenwants at that point, which is whether the elided lines differ from what they already have.
Blocked on #766, deliberately
#766 adds
marker_kindtoledger_foldsand the retrieve-rate query. That query ordersthis work by which shapes actually cost a round trip, instead of by which example someone
remembers. Three wording changes have shipped on reasoning alone; the fourth should not.
Cost to keep in view: any wording that grows the marker reduces fold volume, since the
gate weighs the rendered string. #755 traded 15 bytes on one arm for the identity, and
that trade was never measured either.